Objectives of the service
SEED improves the state of the art in energy forecasting accuracy with geopredict’s proprietary Machine Learning technology using satellite Earth Observation data and High Performance Computing, while also catering to the needs of renewable operators in terms of optimizing solar farm‘s operations and maintenance scheduling based on weather and predictive insights. The SEED system provides forecasting capabilities from short-term to long-term as Software-as-a-Service for the Indian renewable energy market, supporting decision making across a wide range of use cases and applications across different time horizons such as intraday and day-to-day market operations, day-ahead energy production forecasting in 15 minute time resolution, scheduling and deviation settlement mechanism regulations compliance, and multiple weeks to months/years-ahead forecasting of system parameters for operational planning, including predictive maintenance and automated maintenance scheduling.
Users and their needs
SEED is designed to tailor to the needs of Indian renewable operators in terms of optimizing solar farm’s operations and maintenance scheduling based on weather and predictive insights generated by the SEED system and on given state regulations.
The project involves the spectrum of large, medium and small renewable power generation companies:
Mahindra Susten Pvt Ltd is very big player in Solar generation in India with 5 GW installed solar power plants.
AVAADA Energy Pvt Ldt is an important player in Solar generation in India with an installed capacity of 1.5 GW
Pemmasani Solar Pvt Ltd, representing the large amount of smaller renewable generators with and installed capacity below 100 MW.
SEED targets the problem of underperformance of existing renewable energy assets in India, by tackling two main underlying and interlinked issues:
Non availability of location specific and high-resolved weather/climate models tailored to Indian conditions, affecting performance of existing renewable energy forecasting models, and
Lack of effective predictive tools for optimization of solar operations tailored to the Indian context.
The main challenge for geopredict lies in identifying and implementing spatially and temporally high resolved sets of atmosphere earth observation data for the Indian Ocean region.
Service/ system concept
The SEED system is designed to provide predictive information for two purposes to the customer:
Renewable energy production forecast one day ahead in 15-minute time resolution, including 9 intraday revisions
Medium to long term forecasts in daily, weekly, or monthly time resolution for predictive maintenance and operations planning needs.
The SEED AI self-organizing modeling and forecasting engine processes historical and current EO atmosphere data together with required customer energy production data to get optimized energy forecasting models for each customer on site level for power generation forecasting and predictive maintenance. The result is predictive information delivered to the customer in form of numerical data and reports.
Space Added Value
The project massively uses Satellite Earth Observation (EO) as the main source of information for atmosphere modelling and forecasting, which serves the company’s energy forecasting models as input.
For short-term energy forecasting needs defined by India’s Deviation Settlement Management regulation (DSM) EO data with high temporal resolution (15 minutes) are required. Currently, the system uses EO data products from NOOA/GOES. Using these data for DSM-related energy forecasting based on the SEED model has shown a performance increase (penalty decrease) of up to 35% compared to existing solutions. Once the company gets access to India’s geostationary satellite data products, these data will be tested for forecast performance improvement. The same applies when data of EUMETSAT’s Meteosat Third Generation satellite for the Indian Ocean becomes available.
For longer-term forecasts with lower temporal resolution (days, weeks, months, years) for predictive maintenance the project will test public EO data products of COPERNICUS CAMS, CMEMS, and C3S. Here, both surface and pressure level data with a spatial resolution of 0.125° will be considered.
Intense online meetings with the three Indian renewable power generators involved in the project discussing their specific needs and expected improvements for day-ahead Deviation Settlement Management (DSM) forecasting and predictive maintenance, in particular (fig. 1).
The information gained from these meetings is flowing into the preparation of the Requirements and System and Services Architecture Documents. Internal end-user requirements definition will continue and works on DSM implementation, predictive maintenance module, and API implementation for data download and uploads will start next.